| Benchmark | Gemini 3 Pro | GPT-4o |
|---|---|---|
| AA Agentic Index | 52 | 9.7 |
| AA Intelligence | 40.6 | 11.2 |
| AA-LCR | 73 | 55 |
| AA-Omniscience | 15.3 | -10.5 |
| AIME 2025 | 100 | 11.7 |
| arc_agi_1 | 75 | 4.5 |
| ARC-AGI-2 | 54 | 0 |
| arena_vision | 1290 | 1162 |
| Artificial Analysis Coding Index | 46.5 | 24.2 |
| critpt | 9.1 | 0 |
| Fortress | 41.7 | 47.2 |
| gdpval | 34.2 | 0 |
| GPQA Diamond | 91.9 | 70.1 |
| HLE | 45.8 | 5.3 |
| IFBench | 70.4 | 36 |
| LiveCodeBench | 92 | 38.3 |
| LiveCodeBench v6 | 90.7 | 30.9 |
| longbench_v2 | 68.2 | 48.1 |
| mathvision | 86.1 | 30.4 |
| MathVista | 89.8 | 63.8 |
| metr_hcast | 71.3 | 25 |
| metr_re_bench | 35 | 0 |
| metr_swaa | 99.5 | 99.2 |
| MMLU-Pro | 90.1 | 74.7 |
| mmmlu | 91.8 | 81.4 |
| MMMU-Pro | 81 | 59.9 |
| multichallenge | 65.7 | 40.3 |
| MultiNRC | 59 | 12.4 |
| OCRBench | 90.3 | 806 |
| OmniScience Accuracy | 55.8 | 23.7 |
| OmniScience Non-Hallucination | 10 | 62.1 |
| PropensityBench | 52.9 | 46.1 |
| scicode | 56.1 | 33.4 |
| SEAL VISTA | 51.5 | 34.9 |
| simplebench | 76.4 | 17.8 |
| simpleqa | 72.1 | 39.4 |
| swe_bench_bash | 74.2 | 21.6 |
| SWE-bench Verified | 78 | 38.8 |
| Terminal-Bench Hard | 56.9 | 8.3 |
| vectara_answer_rate | 99.4 | 93.8 |
| vectara_avg_summary_length | 101.9 | 86.6 |
| vectara_factual_consistency | 86.4 | 90.4 |
| vectara_hallucination_rate ↓ | 13.6 | 9.6 |
| Video-MME | 88.4 | 77.2 |
| VideoMMMU | 87.6 | 61.2 |
| τ²-Bench Telecom (AA run) | 98 | 28.9 |
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.